728 Part V / Movement
Figure 30–10 Accuracy of movement varies in direct pro-
portion to its speed.Subjects held a stylus and had to hit a
straight line lying perpendicular to the direction in which they
moved the stylus. Subjects started from one of three different
initial positions and were required to complete the movement
within three different times (140, 170, or 200 ms). A trial was
successful if the subject completed the movement within
10% of the required time. Only successful trials were used
for analysis. Subjects were informed when a trial was not
successful. The variability in the motion of the subjects’ arm
movements is shown in the plot as the standard deviation of
the movement endpoint plotted against average speed (for
each of three movement starting points and three movement
times, giving nine data points). The variability in movement
increases in proportion to the speed and therefore to the
force producing the movement. (Adapted, with permission,
from Schmidt et al. 1979.)
10
运动终点标准偏差 (毫米)
200
200
170
140
运动时间(毫秒)
100 150
平均速度(厘米/秒)
50
10
9
8
7
6
5
4
3
2
1
目标线
可能的开始点
movement trajectory or sequence of states that can be
considered optimal. Although noise and environmen-
tal disturbances can cause the motor system to devi-
ate from the desired behavior, the role of feedback is
simply to return the movement back to the desired
trajectory. However, this approach is not necessarily
computationally efficient. Rather than specifying the
desired state of the body, we can specify an optimal
feedback controller to generate the movement.
Optimal Feedback Control Corrects for Errors in a
Task-Dependent Manner
Optimal feedback control aims to minimize a cost
such as a combination of energy and task inaccuracy
(Chapter 34). This type of feedback control is based on
the idea that people do not plan a trajectory given a par-
ticular cost. Instead, the cost is used to create a feedback
controller that specifies, for example, how the feedback
gain for positional errors (and other errors such as veloc-
ity and force) changes over time. Therefore, given the
goal of the task, the controller specifies the motor com-
mand suitable for different possible states of the body.
The trajectory is then simply a consequence of applying
the feedback control law to the current estimate of the
state of the body (Figure 30–11). The feedback controller
is optimal in that it can minimize the cost even in the
presence of potential disturbances.
Optimal feedback control, therefore, does not make
a hard distinction between feedforward and feedback
control. Rather, during a task, the balance between
feedforward and feedback control varies along a con-
tinuum that depends on the extent to which the esti-
mate of current body state is influenced by predictions
(feedforward) or by sensory input (feedback).
An important feature of optimal feedback control
is that it will correct only for deviations that are task
relevant and allow variation in task-irrelevant devia-
tions. For example, when reaching to open an exit door
that has a long horizontal handle, it is of little impor-
tance where along the handle one makes contact, so
deviations in the horizontal direction can be ignored.
Such considerations lead naturally to the minimal
intervention principle that one should only intervene
in an ongoing task if deviations will affect task success.
Intervening will generally add noise into the
system (and require an increased effort), so interven-
ing unnecessarily will lead to a decrement in perfor-
mance. The aim of optimal feedback control is not to
eliminate all variability, but to allow it to accumulate
in dimensions that do not interfere with the task while
minimizing it in the dimensions relevant for the task
completion. The minimal intervention principle is
supported by studies that show that feedback does not
always return the system to the unperturbed trajectory
but often acts in a manner to reduce the effect of the
disturbance on the achievement of the task goal and to
ensure that corrections are task-dependent.
Optimal feedback control emphasizes the setting
of feedback gains, which can be partially instanti-
ated by reflexes that generate rapid motor responses.
Optimal feedback control proposes that these rapid
responses should be highly tuned to the task at hand.
Kandel-Ch30_0707-0736.indd 728 18/01/21 6:00 PM